disable hedging for s3 express (#602)

Author: TingDaoK

source/s3_auto_ranged_put.c

@@ -464,8 +464,11 @@ static void s_s3_auto_ranged_put_send_request_finish(
     struct aws_http_stream *stream,
     int error_code) {
     struct aws_s3_request *request = connection->request;
-    if (request->request_tag == AWS_S3_AUTO_RANGED_PUT_REQUEST_TAG_PART) {
+    if (request->request_tag == AWS_S3_AUTO_RANGED_PUT_REQUEST_TAG_PART && !request->meta_request->is_express) {
         /* TODO: the single part upload may also be improved from a timeout as multipart. */
+        /* Note: For S3express, server responses random ETag. Killing the connection in the middle may cause server
+         * receive two same part and result in confusing during complete MPU with etag doesn't match.
+         * Disable the first byte timeout for s3 express.  */
         aws_s3_client_update_upload_part_timeout(request->meta_request->client, request, error_code);
     }
     aws_s3_meta_request_send_request_finish_default(connection, stream, error_code);
@@ -1216,11 +1219,13 @@ static void s_s3_prepare_upload_part_finish(struct aws_s3_prepare_upload_part_jo
             aws_s3_meta_request_lock_synced_data(meta_request);
             struct aws_s3_mpu_part_info *part = NULL;
             aws_array_list_get_at(&auto_ranged_put->synced_data.part_list, &part, request->part_number - 1);
+            /**
+             * 1. If request retries, get the checksum context from the previous attempt to reuse it. In case of the the
+             *      request body in memory mangled
+             * 2. If this is the first attempt, the context is initialized from the previous prepare step.
+             **/
             AWS_ASSERT(part != NULL && part->checksum_context != NULL);
-            /* Use checksum context if available, otherwise NULL for new parts */
             checksum_context = part->checksum_context;
-            /* If checksum already calculated, it means either the part being retried or the part resumed from list
-             * parts. Keep reusing the old checksum in case of the request body in memory mangled */
             aws_s3_meta_request_unlock_synced_data(meta_request);
         }
         /* END CRITICAL SECTION */

source/s3_checksum_context.c

@@ -182,7 +182,6 @@ struct aws_byte_cursor aws_s3_upload_request_checksum_context_get_checksum_curso
         return checksum_cursor;
     }
     s_lock_synced_data(context);
-    /* If not previous calculated */
     if (context->synced_data.checksum_calculated) {
         checksum_cursor = aws_byte_cursor_from_buf(&context->synced_data.base64_checksum);
     }

source/s3_meta_request.c

@@ -1188,7 +1188,10 @@ void aws_s3_meta_request_send_request(struct aws_s3_meta_request *meta_request,
         options.on_metrics = s_s3_meta_request_stream_metrics;
     }
     options.on_complete = s_s3_meta_request_stream_complete;
-    if (request->request_type == AWS_S3_REQUEST_TYPE_UPLOAD_PART) {
+    if (request->request_type == AWS_S3_REQUEST_TYPE_UPLOAD_PART && !meta_request->is_express) {
+        /* Note: For S3express, server responses random ETag. Killing the connection in the middle may cause server
+         * receive two same part and result in confusing during complete MPU with etag doesn't match.
+         * Disable the first byte timeout for s3 express.  */
         options.response_first_byte_timeout_ms = aws_atomic_load_int(&meta_request->client->upload_timeout_ms);
         request->upload_timeout_ms = (size_t)options.response_first_byte_timeout_ms;
     }